PDF(3641 KB)
人工智能子宫颈细胞病理图像辅助诊断系统的应用：一项前瞻性诊断试验研究
谢玲玲, 叶栋栋, 何贵, 林仲秋, 周晖
中国实用妇科与产科杂志 ›› 2026, Vol. 42 ›› Issue (2) : 212-217.
PDF(3641 KB)
PDF(3641 KB)
人工智能子宫颈细胞病理图像辅助诊断系统的应用：一项前瞻性诊断试验研究
Application of an artificial intelligence-assisted diagnosis system for cervical cell pathology images：a prospective diagnostic test study
目的 通过前瞻性诊断试验研究评估人工智能（artificial intelligence，AI）子宫颈细胞病理图像辅助诊断系统用于诊断子宫颈病变的可行性。方法 收集2022年4月27日到2023年3月21日期间就诊于中山大学孙逸仙纪念医院妇科患者的子宫颈液基细胞学样本347例，采用子宫颈细胞病理图像辅助诊断系统进行AI独立阅片或AI辅助TBS诊断，其结果与专家组及人工独立阅片组进行比较，结合最终组织病理学结果评价前期AI辅助诊断模型的准确性。结果 （1）以专家组TBS诊断为参考标准，AI辅助阅片组与专家组的一致性最高，加权Kappa值达到0.924。（2）以活检或锥切病理结果为参考标准，AI独立阅片组及AI辅助阅片组诊断≥低级别鳞状上皮内病变（LSIL）及≥高级别鳞状上皮内病变（HSIL）的灵敏度均高于人工独立阅片组及专家组（0.614 vs. 0.614 vs. 0.561 vs. 0.579，0.769 vs. 0.590 vs. 0.564 vs. 0.513）。（3）AI诊断系统联合人乳头瘤病毒（HPV）筛查检出≥HSIL病变的准确率为75.3%，高于人工独立阅片组（67.1%）及参考组（74.0%）。结论 该诊断系统用于子宫颈TBS分类有较高的准确性及稳定性，对子宫颈病变可能具有较人工更高的预测价值。
Objective To assess the feasibility of an artificial intelligence （AI）-assisted diagnosis system for cervical cell pathtology images in diagnosing cervical lesions through a prospective diagnostic text study. Methods A total of 347 liquid-based cytology samples were collected from patients who visited the Department of Gynecology at Sun Yat-sen Memorial Hospital，Sun Yat-sen University between April 27，2022 and March 21，2023. The samples were independently analyzed by the AI system or with AI assistance in ThinPrep Bethesda System （TBS） classification. The results were compared with those obtained from the expert group and the pure researcher-independent reading group. The accuracy of the AI-assisted diagnosis model was evaluated by taking histopathological findings into account. Results Using the TBS diagnosis of the expert group as the reference standard，the highest consistency was observed between the AI-assisted reading group and the expert group，with a weighted Kappa value of 0.924. When biopsy or conization pathological results were used as the reference standard，the sensitivity of the AI-independent reading group and the AI-assisted reading group in diagnosing ≥LSIL and ≥HSIL diseases was higher than that of the pure researcher-independent reading group and the expert group （0.614 vs. 0.614 vs. 0.561 vs. 0.579；0.769 vs. 0.590 vs. 0.564 vs. 0.513）. The accuracy of the AI diagnosis system combined with HPV screening in detecting ≥HSIL lesions was 75.3%，which was higher than that of the pure researcher-independent reading group （67.1%） and the expert group （74.0%）. Conclusion The AI diagnostic system demonstrates high accuracy and stability in cervical TBS classification, and may possess superior predictive value for cervical lesions compared with manual diagnosis.
人工智能 / 子宫颈细胞学诊断系统 / 子宫颈癌筛查
artificial intelligence / cervical cytology diagnosis system / cervical cancer screening
| [1] |
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
Computer-assisted diagnosis is key for scaling up cervical cancer screening. However, current recognition algorithms perform poorly on whole slide image (WSI) analysis, fail to generalize for diverse staining and imaging, and show sub-optimal clinical-level verification. Here, we develop a progressive lesion cell recognition method combining low- and high-resolution WSIs to recommend lesion cells and a recurrent neural network-based WSI classification model to evaluate the lesion degree of WSIs. We train and validate our WSI analysis system on 3,545 patient-wise WSIs with 79,911 annotations from multiple hospitals and several imaging instruments. On multi-center independent test sets of 1,170 patient-wise WSIs, we achieve 93.5% Specificity and 95.1% Sensitivity for classifying slides, comparing favourably to the average performance of three independent cytopathologists, and obtain 88.5% true positive rate for highlighting the top 10 lesion cells on 447 positive slides. After deployment, our system recognizes a one giga-pixel WSI in about 1.5 min.© 2021. The Author(s).
|
| [11] |
Technical advancements significantly improve earlier diagnosis of cervical cancer, but accurate diagnosis is still difficult due to various factors. We develop an artificial intelligence assistive diagnostic solution, AIATBS, to improve cervical liquid-based thin-layer cell smear diagnosis according to clinical TBS criteria. We train AIATBS with >81,000 retrospective samples. It integrates YOLOv3 for target detection, Xception and Patch-based models to boost target classification, and U-net for nucleus segmentation. We integrate XGBoost and a logical decision tree with these models to optimize the parameters given by the learning process, and we develop a complete cervical liquid-based cytology smear TBS diagnostic system which also includes a quality control solution. We validate the optimized system with >34,000 multicenter prospective samples and achieve better sensitivity compared to senior cytologists, yet retain high specificity while achieving a speed of <180s/slide. Our system is adaptive to sample preparation using different standards, staining protocols and scanners.
|
| [12] |
Cervical adenocarcinoma is a heterogenous group of tumours with various aetiologies, molecular drivers, morphologies, response to treatment and prognosis. It has become evident that human papillomavirus (HPV) infection does not drive all adenocarcinomas, and appropriate classification is critical for patient management, especially in the era of the HPV vaccine and HPV-only screening. Identified as one of the most important developments in gynaecological pathology during the past 50 years, the separation of cervical adenocarcinomas into HPV-associated (HPVA) and HPV-independent has resulted in a transformation of the classification system for cervical adenocarcinomas. HPVA has been traditionally subclassified by morphology, such as usual type (UEA), mucinous and villoglandular, etc. However, it has become evident that cell type-based histomorphological classification is not clinically meaningful, and the newly proposed International Endocervical Adenocarcinoma Criteria and Classification (IECC) is a necessary and relevant break from this prior system. Non-HPV-associated adenocarcinomas can be divided by their distinct morphology and molecular genomics with very different responses to standard therapies and potential for future targeted therapies. These include gastric-type, clear-cell, mesonephric and endometrioid adenocarcinomas. So-called 'serous' carcinomas of the cervix probably represent morphological variants of UEA or drop metastases from uterine or adnexal serous carcinomas, and the existence of true cervical serous carcinomas is in question. This review will discuss the advances since WHO 2014, and how HPV status, pattern of invasion as described by Silva and colleagues, histological features and molecular markers can be used to refine diagnosis and prognostication for patients with cervical adenocarcinoma.© 2019 John Wiley & Sons Ltd.
|
| [13] |
|
| [14] |
Background: Data on the current burden of adenocarcinoma (ADC) and histology-specific human papillomavirus (HPV) type distribution are relevant to predict the future impact of prophylactic HPV vaccines.;Methods: We estimate the proportion of ADC in invasive cervical cancer, the global number of cases of cervical ADC in 2015, the effect of cervical screening on ADC, the number of ADC cases attributable to high-risk HPV types -16, -18, -45, -31 and -33, and the potential impact of HPV vaccination using a variety of data sources including: GLOBOCAN 2008, Cancer Incidence in Five Continents (CI5) Volume IX, cervical screening data from the World Health Organization/Institut Catala d'Oncologia Information Centre on HPV and cervical cancer, and published literature.;Results: ADC represents 9.4% of all ICC although its contribution varies greatly by country and region. The global crude incidence rate of cervical ADC in 2015 is estimated at 1.6 cases per 100,000 women, and the projected worldwide incidence of ADC in 2015 is 56,805 new cases. Current detection rates for HPV DNA in cervical ADC tend to range around 80-85%; the lower HPV detection rates in cervical ADC versus squamous cell carcinoma may be due to technical artefacts or to misdiagnosis of endometrial carcinoma as cervical ADC. Published data indicate that the five most common HPV types found in cervical ADC are HPV-16 (41.6%), -18 (38.7%), -45 (7.0%), -31 (2.2%) and -33 (2.1%), together comprising 92% of all HPV positive cases. Future projections using 2015 data, assuming 100% vaccine coverage and a true HPV causal relation of 100%, suggest that vaccines providing protection against HPV-16/18 may theoretically prevent 79% of new HPV-related ADC cases (44,702 cases annually) and vaccines additionally providing cross-protection against HPV-31/33/45 may prevent 89% of new HPV-related ADC cases (50,769 cases annually).;Conclusions: It is predicted that the currently available HPV vaccines will be highly effective in preventing HPV-related cervical ADC.
|
| [15] |
High-risk human papillomavirus (HR-HPV) testing is more sensitive than cytology for the detection of cervical cancer and its precursors. However, limited and inconsistent data are available about the efficacy of the combination of these two methods for screening cervical adenocarcinoma. This multicenter retrospective study investigated the screening results of a cohort of Chinese patients who were subsequently diagnosed with invasive cervical adenocarcinoma, with the goal of identifying the optimal cervical adenocarcinoma screening method.
|
| [16] |
马德勇, 冯慧, 贺丹, 等. 子宫颈细胞学非典型腺细胞临床价值分析[J]. 中国实用妇科与产科杂志, 2022, 38(3):331-335.DOI:10.19538/j.fk2022030117.
|
| [17] |
| [18] |
冯慧, 赵撼宇, 赵健, 等. 人工智能识别阴道镜下子宫颈红区在子宫颈癌前病变诊断中的价值[J]. 中国实用妇科与产科杂志, 2025, 41(3):357-360.DOI:10.19538/j.fk2025030120.
|
利益冲突 所有作者均声明不存在利益冲突
/
| 〈 |
|
〉 |
